Mining Moving Object Trajectories Outliers And Association Patterns

Along with the advance of technology, various services are increasing for moving objects.Moving object trajectory data as historical activity data of moving objects, in some degree can reflectsome internal and external features of moving objects, such as properties, status, behaviors and so on.This paper separately carries on the related research in view of the two kinds of moving objecttrajectories: normal trajectories and abnormal trajectories based on anomaly detection and correlationanalysis in data mining technology. The Main work is as follows:(1) Abnormal trajectory detection method based on trajectory partition.In view of the existing anomaly detection method cannot detect the abnormal substrajectories,this paper proposes a trajectory abnormal detection algorithm based on trajectory partition(Partition_Detection). This method can be divided into two stages. Firstly each trajectories should bedivided into several substrajectories using two granularities, in order to ensure effectiveness of thesegmentation results and efficiency of the algorithm, which would be used for the next phase--anomaly detection. Then this paper introduces the notion of adjustment coefficient on the basis ofdistance based anomaly detection method, and proposes a detection method combining density anddistance. This method could improve the quality of abnormal trajectory detection, and avoid theoccurring of the dense areas missing.Experimental results show that the Partition_Detection algorithm can successfully detectabnormal subtrajectories, and abnormal trajectories in dense areas. In addition, the algorithmperformance is satisfactory. The partition strategy with two granularities can prune off a lot ofcomparisons between coarse-grained subtrajectories, greatly improving efficiency of the algorithm.(2) Trajectory association pattern mining method based on the AprioriAll algorithm.After eliminating abnormal trajectories, association patterns the rest trajectories would be minedby using correlation analysis algorithm in data mining. Firstly this paper proposes an associationpattern mining algorithm (Pattern_Mining) based on the AprioriAll algorithm, which is simple-logical,clear, effective and easily applying for trajectory dataset. According to the predetermined minimumsupport and confidence thresholds, all frequent trajectory sequences with different length would bemined, and their corresponding association rules would be generated. Secondly on the basis of thePattern_Mining algorithm, this paper proposes an incremental updating association pattern miningalgorithm (Pattern_Mining_UP). The algorithm makes full use of the original frequent trajectory setsand only mines a little part of data set, to avoid the redundant operations and improve the miningefficiency.Experimental results show that the Pattern_Mining algorithm can mine trajectory associationpatterns properly. In addition, the incremental algorithm (Pattern_Mining_UP) is also satisfactory.When the data set increases rapidly, the incremental mining can greatly improve the efficiency, andreduce the running time.